THE INNER CIRCLE GUIDE TO SPEECH ANALYTICS. Sponsored by

Transcription

1 THE INNER CIRCLE GUIDE TO SPEECH ANALYTICS Sponsored by Published July 2011 I

2 CONTENTS About GemaTech... 3 About ContactBabel... 3 The Technology of Speech Analytics... 5 The Business Benefits of Speech Analytics Compliance Agent evaluation and improvement Contact centre performance improvements Business process improvements Business Intelligence Improving the Customer Experience Increasing Profitability Building a Business Case for Speech Analytics Estimating ROI Inhibitors to speech analytics The Implementation and Use of Speech Analytics Pre-implementation: Selecting a solution The implementation process Post-Implementation: Using Speech Analytics The Market Landscape of speech analytics Business drivers Key verticals and activities Pricing Company Profile Future Directions for Speech Analytics

3 ABOUT GEMATECH The Ultimate Live Speech Analytics Solution Deliver 100% compliance on every call with GemaTech s new Call Analyzer. Analysis of every call, as it is taking place, provides instant feedback to individual agents, supervisors and managers to ensure that all agents are following their scripts correctly. Desktop prompts are delivered to the agent s screen during calls reminding them if they have forgotten to say necessary phrases, to achieve first call resolution and to ensure legally binding, FSA compliant, contracts on every call. Call Analyzer combines the power of speech to text, phonetics and key phrase search techniques to deliver 90 to 97% accuracy on 100% of calls. Bespoke management reports deliver fast business insight in any format. Quick implementation and low set up costs deliver a fast return on investment, with a proof of concept trial available if required. Contact: Karen Jones T: +44 (0) E: k.jones@gematech.com A: Telford House, Hamilton Close, Houndmills, Basingstoke, Hampshire RG21 6YT (UK) W: ABOUT CONTACTBABEL If you have a question about how the contact centre industry works, or where it s heading, we have the answer. Our major ongoing primary research projects match our experience analysing the contact centre industry. We understand how technology, people and process work together, and what their future holds. We help solution providers develop their marketing strategies and talk to the right prospects. We've shown governments how the global contact centre industry will change and affect their nation. We help contact centres understand how to improve, and what their customers think of them. If you have a question about your company s future in the contact centre industry, we can help you. 3

4 Call Centres. First call resolution can be a reality for every call. When calls have to be right first time, you need GemaTech s Live Call Analyzer Faster: 100% of calls analysed live including live agent prompting with fast, easy implementation. Better: 90-97% accuracy on 100% of calls, delivering first call resolution and live compliance monitoring. Cheaper: Low license charge per agent and cloud-based option reduces cost per call and sale. Reduces: Risk of non-compliance, call abandon and call attrition rate. GemaTech (UK) Ltd, Telford House, Hamilton Close, Houndmills, Basingstoke, Hampshire RG21 6YT. Tel: +44 (0) Fax: +44 (0) info@gematech.com GemaTech s Call Analyzer Powered by Elsbeth Vocal Coach +44 (0) The Telecoms Continuity Specialists

5 THE TECHNOLOGY OF SPEECH ANALYTICS In the late 1990s, data warehousing was a big growth industry, especially in sectors such as retail, where the widespread usage of customer loyalty cards gave huge amounts of data about customers, their buying patterns and preferences. However, getting the data into storage was not the difficult bit: the greatest value came from being able to identify and analyse the relevant and insightful patterns within these data, through data mining. In many cases, the reality never lived up to the hype, as the analytical capabilities of data mining tools and businesses' ability to use them effectively did not match the ease with which the data warehouse was filled in the first place. Speech analytics solutions are analogous with the data warehousing and mining applications in as far as they analyse huge quantities of data - here, call recordings - and identify important and insightful patterns in caller and agent activity. Hence, speech analytics also called audio mining. (It should be noted that some speech analytics solutions act in real-time, so the analogy is not quite exact). However, unlike the gap in functionality between data warehousing and data mining that we saw a decade ago, speech analytics solutions offer a proven and insightful option to release the customer value that is stored in these enormous quantities of information: insight about the customer, the agent, the business processes and the products and services that the business sells. Like most contact centre applications, speech analytics can be used to cut costs, but its promise goes far beyond this. No other contact centre technology provides the business with this level of potential insight that goes far beyond the boundaries of the contact centre, and can offer genuine and quantifiable ways in which sub-optimal business processes can be improved. This is not to say that speech analytics is at its zenith. Significant improvements can be made to the accuracy and speed of the speech engines, the sophistication of analytical capabilities and the usability of reports. Some of the actionable findings from speech analytics may seem very simple - the recommendation to change a few words in a script, for example - but the potential impact upon the cost, revenue, agent capability and customer experience that is possible through speech analytics is perhaps unprecedented. Of course, it is not enough to plug in the technology and wait for results. The most important element to getting the most out of speech analytics is to use the application properly: ask the right questions, listen to the right calls, make the right decisions and get the right people to support and act upon findings. 5

6 The elements of speech analytics There are various elements to speech analytics solutions, including: Speech engine: a software program that recognizes speech and converts it into data (either phonemes - the sounds that go to make up words - or as a text transcription). Indexing layer: a software layer that improves and indexes the output from the speech engine in order to make it searchable Query and search user interface: the desktop application where users interact with the speech analytics software, defining their requirements and carrying out searches on the indexed data Reporting applications: the presentation layer of speech analytics, often in graphical format Business applications: provided by vendors, these pre-defined modules look at specific issues such as adherence to script, debt collections etc, and provide suggestions on what to look for. Phonetic or Speech-to-Text (LVCSR) Speech analytics solutions use speech engines that are either phonetic or speech-to-text / LVCSR (Large Vocabulary Continuous Speech Recognition). In LVCSR, the call is converted into text in order for analysis to take place, and depend upon a language model and dictionary to identify words correctly. Phonetics-based applications - which look for defined sounds or strings of sounds - attempt to match these sounds to target words or phrases in a phonetic index file. The phonetics approach does not require a language model or predefinition of every word looked for, meaning indexing and searching is more rapid. However, LVCSR approaches can offer a greater understanding of what is in each call. Vendors that use speech-to-text engines point out that even if a phonetic system accurately identifies a key word, that there is no guarantee that it will be used in the correct context (for example, the word "website" may not just refer to a company's own site, but to a competitor's or something else entirely). Phrase recognition (such as 'the website doesn t work', or similar) can be used to alleviate this, although the number of true positive results using this method can be lower, as there are far more ways to say a similar thing. Solution providers that offer analytics based upon a phonetic speech engine state its usefulness where customers already know the type of words and phrases that they are looking for, based on their business needs. Many vendors also have significant experience with specific business sectors and call types, and can offer useful advice on how to maximize the volumes of data available for analysis. 6

7 Figure 1: Some advantages and disadvantages of LVCSR/speech-to-text and phonetic speech analytics solutions LVCSR / speech-to-text Phonetic Advantages A complete transcript of the call is available for detailed analysis and viewing Ability to carry out 'discovery', uncovering trends or events which an organization may not be aware of Fast text-based search Level of analysis possible tends to be deeper than phonetic-based solutions Faster indexing time Quick and easy to maintain, once the customer knows what they are looking for Tends not to require a dedicated employee Initial deployment in days No requirement to define words and phrases upfront Not dependent on dictionary or language model Disadvantages Slower indexing time (25-30x real-time) than phonetic systems Low accuracy initially (most LVCSR systems also use some phonetic and/or phrase-based technology as well) Word recognition is dependent upon it being in the dictionary within the language model, which requires updates Longer time to implement and ongoing fine-tuning No guarantee that an identified keyword will be used in the right context Slower searches than LVCSR-based systems Homophones and homonyms produce false positives Conversations not viewable by the end-user as text Measurements of accuracy Speech-to-text solutions are measured by the word-error rate: how many words are incorrectly identified? Yet identification of even less than 50% of words is often enough to provide a solid base of data upon which to perform analysis. A speech-to-text transcript of a conversation can appear wildly inaccurate to the reader, yet will often provide enough accurate reference points and keywords upon which to perform complex and insightful analysis. Potential customers should be aware that there is far more to a successful speech analytics solution than getting close to 100% accuracy for word recognition. Phonetic solutions' measurements are made up of precision (or accuracy) and recall (or detection). As an example, if there are 100 files searched for specific words, which occur in 60 of them, then if there are 30 'hits' returned - all of which contain the word or phrase - that is measured as 100% precision, and 50% recall. 7

8 Solutions can be set at a certain confidence level (i.e. confidence that there will be no more than x% of results as false positives or negatives), depending on the business need, as some issues, such as compliance, require very high confidence levels to be maintained. When considering which solution to implement, customers should ask not only about the accuracy of the solution, but also about the recall, detection or completeness rates. The call recording environment also has a significant part to play in these results, as digitally-recorded, stereo/dual channel recordings will provide more opportunities for the speech and analytics engines to identify words and phrases correctly. Real-time analytics There is some debate amongst vendors upon who can provide true real-time analysis, and potential customers for whom this is an issue would do well to investigate this area fully. Some vendors take a parallel feed, streamed live into their solution and can act upon this within the call, which is especially useful for compliance and for forming legally-binding contracts on the phone, where specific terms and phrases must be used and any deviation can be flagged to the agent's screen on the call. Getting these calls right first-time obviously impacts positively upon first-call resolution rates, and through picking up phrases such as "speak to your supervisor", can escalate calls automatically. Real-time offers a big step up from the traditional, manual call monitoring process, and offers real-time reporting on compliance as well. Finance, telecoms and utilities companies - and indeed, any business where telephone-based contracts are important - are particularly interested in this. Very high levels of accuracy are vital for this type of speech analytics. Not all vendors are pursuing this aggressively, with some preferring to concentrate their R&D efforts on tighter integration with their WFO suite, expanding speech analytics into multichannel and offering deeper analytical insight on masses of call recordings. However, for businesses which need immediate feedback within the call, true real-time analysis is certainly available today, and R&D efforts are being stepped-up to link real-time analytics with the CRM systems already in place, in order to tailor offers to customers based on what is being said within the call. 8

9 --- Advertorial --- Get it right before you put the phone down! Speech analytics which coaches your agents during every call is the only way to deliver first call resolution before your agent puts the phone down. Call centres tend to suffer from similar problems: Can they be sure that their agents are always providing a high level of customer service and being polite and helpful to customers? How can they be sure that agents are not mis-selling products? Are agents forgetting to say certain key phrases? Therefore call centre managers need to be sure that their strategy for improving customer service, achieving compliance and increasing first call resolution rates is going to deliver and achieve a demonstrable Return on Investment (RoI). The latest innovation in speech analytics delivers the ultimate in phrase recognition which satisfies all these criteria. As a call centre manager think of what your ideal speech analytics solution would be I am guessing that it would be capable of listening to every call that is handled by your agents all day, every day, and I m also guessing that you would like fast analysis on those calls which provides you with feedback in real-time on the calls which have issues? You may even be contemplating the possibility of using that analysis to help the agent or customer determine the outcome of the call before they put the phone down rather than later or the next day when the customer may not be available for a call back. This type of solution is now a reality not a wish list. The latest innovation in speech analytics has the ability to analyse 100% of calls live, by searching for and highlighting pre-established phrases which need to be said or indeed not said, and consequently delivering a pop-up on the agent s screen which prompts them to: say certain phrases, highlight and score them down for bad language, tell them they are speaking too loud or fast or simply to confirm that they have said all that is required in accordance with their script or call flow. Just think for a second of the many opportunities this new technology creates.rather than making management decisions based on a sample of calls, you can take data from every call and make informed decisions on a true reflection of your call centre, not what you might think is the truth. What could your marketing department do with live market data from customers and prospects about attitudes to you or your competitor s products? How could that insight quickly and positively impact marketing campaigns? If you are an FSA regulated call centre think about the reduction in risk and the huge amount of savings you could make in fines and compensation by automatically monitoring all calls for compliance. Think of the overhead savings you could make in eliminating manual call monitoring and evaluation If you could be on every call your agent handles (figuratively speaking) and be able to intervene for the benefit of your customers, how quickly would your first call resolution rates and bottom line revenue figures start to improve? Don t be satisfied with sampling and post-call analytics, real-time analysis is the key and available now. For further information about this new type of speech analytics technology, please contact GemaTech on +44 (0) or me at k.jones@gematech.com.

10 Accents and dialects In businesses with multiple global operations, speech analytics solutions will of course require different searches and dictionaries for each language, but it is possible to unify reporting across languages if required. If a business has multiple contact centres speaking the same language but with very different accents (for example UK English and US English), it is possible to use the same language model. However, for accents which are very different and has its own cadences and rhythms - for example, Indian English - a different language model may be required, although all the audio can be analysed centrally within the same application. End-user question: "What is the ability of the technology to handle the different languages and accents in a multi-cultural community / country?" (MD, Canadian outsourcer) The technology is now so advanced that in excess of 40 different languages can be recognised including Chinese and Japanese. The ability to discern words and phrases spoken in different regional dialects is also available. GemaTech s Call Analyzer has been proven to achieve 97% accuracy rates on 100% of calls and has been tested on strong accents such as Scottish and English speaking German accents with very accurate results. In addition, tweaking of the configuration for accents is so easy to do that the customer can do this themselves if required. 10

11 THE BUSINESS BENEFITS OF SPEECH ANALYTICS Most contact centre solutions have a specific, easily-communicated reason for purchase, usually around cost savings. The most popular and widespread solutions, such as IVR, workforce management, CTI and outbound dialling, have all had a clear and quantifiable route to cost savings and improved efficiency. Speech analytics has a different appeal to contact centres, and can be used in many different ways to address various business issues. This is an advantage - it is hugely flexible - but it can also make its message to the market more complicated, and to the cynical, it can seem as though speech analytics is claiming to solve every problem that a contact centre could possibly have. However, depending upon how speech analytics is used, it can certainly assist in cost reduction, agent improvement, business process optimisation, avoidance of litigation and fines, customer satisfaction and loyalty improvements, and increases in revenue. COMPLIANCE Many businesses, especially those in finance, insurance, public sector and debt collection, have become encumbered with regulations which they must follow strictly, with potentially expensive penalties for failure, including heavy fines and criminal prosecution. Contact centres have tried to reduce their risk through scripting, call monitoring and call recording, but these do not offer any guarantees or proof of compliance. Speech analytics means that 100% of calls can be verified as compliant - and be proven to be so - preventing disputes or escalation of enquiries by monitoring the exact language used within each call. Return on investment comes from the avoidance of litigation and fines, and the use of speech analytics for compliance is very prevalent, especially in North America. AGENT EVALUATION AND IMPROVEMENT Improve the quality monitoring programme Speech analytics tries to takes the guesswork out of improving customer experience, agent performance and customer insight. By moving from anecdotal or fact-based decisions, from qualitative to quantitative information, some order is put on the millions of interactions that many large contact centres have in their recording systems, improving the reliability of the intelligence provided to decision-makers. It doesn t remove the need to listen to calls, but it means that the calls listened to are far more likely to be the ones that should be listened to, whether for agent evaluation or business insight. Customers using speech analytics can carry out an evaluation of chosen calls - for example, unhappy customers - the results of which can be then be fed back into the existing quality assurance process. This can take the same existing path, without upheaval or any need for altering the QA/QM process, only improving the quality and accuracy of the data used by the existing solution. 11

12 End-user question: "How does the addition of speech analytics impact staffing does it require more people to capture and analyze results OR less because the quality process is more automated requiring less manual call evaluation?" (US outsourcer) New speech analytics solutions such as Call Analyzer require significantly less staff to monitor calls. Because the analysis is fully automated there is no need for any employee to listen regularly to calls. All you need is one or maybe 2 people, depending on the size of call centre, to configure the search criteria and take the results of the analysis to feed them back into the business. In results achieved in Germany to date an average of 30-70% saving on staffing requirements is the norm with the average of 50% being consistently achieved. Identify agent training requirements Apart from 100% monitoring of calls, speech analytics is used to flag cases of talk-over, as well as silence detection. The former can be a source of irritation to the customer and long silences can indicate lack of agent knowledge, although long system navigation times or delays in system response times can also cause this. The analysis of these types of call will identify which of these issues is really the problem. Cut new-starter attrition rates Additionally, speech analytics will also make the training and coaching received by new agents in particular far more effective and targeted. This is especially important for this class of agent, as many operations report that half of their overall staff turnover occurs in the first 90 days of the job, when agents are obviously less-skilled or confident about their role or the organization. Speech analytics can identify the types of behavior - good and bad - that lead to successful call resolution or otherwise, and these can be presented in a targeted way to the new agent to fast-track them to a level of competency that should reduce attrition based on a feeling that they simply can't do the work to a high-enough quality. 12

13 CONTACT CENTRE PERFORMANCE IMPROVEMENTS On first glance, speech analytics can be seen as providing similar information to management information and reporting systems - taking masses of data and making sense of what they mean to the contact centre's performance and perhaps even inside the wider business. However, the vital thing to understand about speech analytics is that it gives contact centres the answer to 'Why', not just 'What'. Why are average handle times so different across agents? Why are customers of this product upset? Why are people calling the contact centre? With high quality data inputs, mixing audio information with data such as call outcomes and revenues, analytics also identifies patterns which the business had no idea even existed, suggesting best practice and identifying areas for improvement at agent, contact centre and process levels. There are numerous possibilities for how speech analytics can impact upon some of the key performance indicators of the contact centre, whether sales- or service-focused, inbound or outbound. Why are customers calling? No other contact centre solution can provide a solid understanding of why customers are calling. Categorising types of calls, and then analyzing them for the occurrence of similar types of words and phrases can give an insight into the reasons for customers' calls. For example, a category such as 'sales' might be analysed for patterns, and it is discovered that the words 'delivery' and 'website' are mentioned in a disproportionate number of them. Listening to some of these conversations, it may be found that the website does not highlight delivery times effectively enough, leading to unnecessary calls to the contact centre, rather than the customer purchasing on the website. Call transfers Rather than making an agent use a call disposition code when they pass a call to another agent (which they may forget to do, or code inaccurately), speech analytics can identify the reasons for passing calls to other agents and putting customers on hold (whether lack of training, broken processes or lack of access to the right systems). 13

14 First-call resolution A major metric for contact centre and customer experience success, first-call resolution can be increased by identifying repeat callers and eliminating the root cause of repeat calls. An example of this was a government institution where they had identified repeat issues as being a problem. Analysing the calls categorized as such, it was found that agents were saying "we'll call you back within 3 hours". As the callers were very keen to get the issue resolved, they were prone to overestimate the time passing, so analysis found that many called back before the three hours were up. By changing the script to e.g. "It's now 11.45am, we'll call you back by 2.45pm", customer expectations were set and call-backs dropped immediately. A few weeks later, call-backs went back up, and it was found that many agents had gone back to the 'old ways', and had forgotten to give the exact time. Average handle time Average call duration / average handle time has traditionally been one of the main measures of a contact centre's 'success', at least when judged by those outside the operation whose focus has often been on cost reduction. In recent years, an increasing focus on the customer experience and first-call resolution has meant that AHT is viewed as less important than previously. However, almost every contact centre still tracks this as a metric, as it is closely linked with cost and performance. Long call durations may be linked with poor agent abilities, lack of knowledge, navigation between systems or very complicated calls, and of course, impact on cost, queue times and the customer experience. Short AHTs can be as bad, if not worse, as they can indicate lack of agent capabilities (so agents pass the call to a colleague, or even deliberately lose the connection), that the contact centre is handling too many simple calls that might be better handled by self-service or that there is a quick and easily-resolved common issue, the solution to which could be propagated in the IVR announcement, on the website or via /sms. The problem for businesses is that they often don't know with any level of confidence why call durations differ. Speech analytics allows businesses to categorise each type of call, and through root-cause analysis, determine what a reasonable length for each type of call is, and investigate the outlying anomalies, either on an agent level, or more widely, by comparing the amount of time taken on each category of call now compared to the past. The identification of calls resolved successfully in a reasonable amount of time will also provide the training department with examples of best practice. 14

15 BUSINESS PROCESS IMPROVEMENTS Everyone connected with the contact centre industry has always known that there is huge insight and knowledge held within the operation and its agents, but which has never before had the ability to be quantified or acted upon by the wider business. Speech analytics offers the ambitious business the greatest potential for improvements in business processes, but there is a great danger of underachievement with so many departments and divisions potentially involved. In the course of researching this report, we have found that the marketing and website departments are the non-contact centre areas most likely to be benefiting currently from insights about customers' views, but there are also examples of how delivery, provisioning, billing and even warehousing departments have learned from the analysis of customers' experiences in the contact centre. The quality of insight and its actionability is totally dependent on a swift reporting process, simple yet rich intelligence, the ownership of process improvement at senior level and before/after comparisons to prove success. Cross-department rivalries or poor communication are a real risk to this. BUSINESS INTELLIGENCE Customer Satisfaction Surveys There has been a great increase in customer satisfaction surveys in recent years, with the widespread uptake of Net Promoter being a good example of companies' desire to learn what their customers actually think about them. However, research has shown that a 'satisfied' customer isn t necessarily a profitable or loyal one, and the results of customer surveys, particularly the written or telephone-based variety (the latter of which, despite its limitations and expense, is still seen as the best method), are carried out at a time when any feelings about the original interaction may have changed or dissipated, are prone to inaccuracy, delay and lack of detail. With all of the methods of customer surveys, the questions are fixed in advance, and if the right questions aren t asked, the level of actionable insight is low. In many cases, a business might know that x% of its customers are satisfied, and y% dissatisfied, but it still has no real idea why this is, or even how it will impact upon their profitability. As an alternative to customer satisfaction surveys, speech analytics allows a business to gather customers' views within the interaction itself - guaranteeing immediacy and accuracy - and can be applied across 100% of calls, rather than focusing on the outlying 'very dissatisfied' or 'delighted' customers. Furthermore, through widespread and detailed analysis of what the call is about, the type of language or messages used in the call, how the customer was handled, and the eventual outcome, businesses will be able to learn how to improve their customer retention and satisfaction in real-life, by-passing the standard metric (e.g. "83% of customers are satisfied") and getting to the root causes of satisfaction or dissatisfaction and sharing the results with the rest of the operation. 15

16 Customer Insight As introduced above, one of the greatest advantages that speech analytics can provide is the ability to understand why things are happening, rather than just what is going on. With many solutions, it is not even necessary to know what you are looking for: automatic categorisation of calls into their constituent types is a starting point, based on the types of words and phrases that typically get used within these types of calls (e.g. "complain", "not happy", "disappointed", "speak with a manager" etc, will often relate to customer complaints). Non-audio data, such as the activity of account closure, refunds etc can also be captured from the screen and linked with the call to provide richer data for analysis. The tracking of word usage compared with its historical use (e.g. a 300% rise in the use of the phrase "can't log-on" after a software upgrade) can quickly indicate and identify issues that can be handed to the relevant department much more quickly than typical inter-department channels could usually manage. Regular references to competitors and their products can be captured, analysed and passed to the marketing or pricing teams to provide them with real-life, rapid and accurate information upon which to base decisions. Crisis management and reaction A solution with automated root-cause analysis capabilities - constantly looking for anomalies and new patterns - can identify spikes in unusual activity shortly after it happens, alerting specific users to the key issues so as to handle them before it runs out of control, damaging brand or customer satisfaction. Product and pricing feedback Speech analytics allows businesses to seek out key words and phrases, such as competitors' names or any instances of pricing, or to gather feedback after a marketing campaign goes out. 16

17 End-user comment: "It is too expensive to put numbers behind what you already know because your agents can tell you." (Head of Reservations in a UK travel company) Can you be sure that your agents are being 100% truthful, 100% of the time? If you consider how long it would take to interview every agent about every call they handle surely this time would be better spent keeping the agents on the phone? Next generation speech analytics solutions can listen to every call and report back in real-time, giving accurate and fast analytical data which does not impact upon call handling time. Also if you can trust your agents to flag up everything you need to know then surely there is no need ever to monitor, record or evaluate calls, let alone analyse them? Secondly if agents forget to say particular phrases for compliance purposes, they are not likely therefore to flag up the fact every time they forget to do so. Implementing real-time speech analytics, such as Call Analyzer from GemaTech, is not just about delivering data on the calls in your call centre, it is more about correcting human error during the call to deliver right-first-time and improved customer service, therefore improving efficiency and increasing revenue in the long term. Example: A major international hotel group ran a campaign calling for people over the age of 30 to attend an event. The call agents were commissioned on getting numbers of people to the event. On the day of the event many who turned up were under the age of 30 and did not qualify. The agents clearly did not qualify their candidates and had no motivation to flag this up to their management. 17

18 IMPROVING THE CUSTOMER EXPERIENCE Factors that impact the customer experience - such as first-call resolution and shorter call and queue times - have been addressed already. This section looks at the handling of complaints, and how speech analytics can take into account the entire customer experience outside the contact centre. Complaints handling Complaints are a potentially rich environment for businesses to understand where they are going wrong, and which issues are in danger of turning a customer into an ex-customer. For many businesses, each complaint is dealt with on a case-by-case basis, with little in the way of categorization or structure being put in place formally, and little chance of communicating findings in an actionable way to the relevant department. Speech analytics gives businesses a chance to quantify the reasons that customers complain, identifying the most important factors, assessing trends and spikes, and providing hard recommendations based on every call taken. 4% of UK calls and 8% of US calls received by contact centres are complaints, with respectively 87% and 80% of these being about problems elsewhere in the enterprise (rather than in the contact centre). Understanding and acting upon what is driving these complaints will clearly make a huge difference to cost and customer satisfaction. On an individual-call basis, real-time analytics allow businesses to track words and phrases related to complaints (such as 'supervisor', 'manager', 'complain', 'unhappy' etc.), allowing escalation to a supervisor, or screen-pop to the agent to provide them with a revised script or suggestions of how to handle the call. Emotion detection may also be used to identify these customers. End-user question: "What alerts are available - for example a customer decides to hang up part way through - how is this managed?" (UK outsourcer) This will remain the same as it is managed currently. All good existing call centre technology has the ability to highlight when calls are prematurely disconnected, as agent scores will provide this information. Notes would be made about the call in the agent s CRM system stating that the customer had hung up. The difference with live speech analytics is that being able to search in real-time on bad phrases or specific words, the reasons for why the customer hung up can be highlighted and addressed which should bring down the re-occurrence of abandoned calls on an ongoing basis. 18

19 The customer experience outside the contact centre There is an increasing requirement and interest in multichannel analytics, including considering , text chat, IVR and web browsing sessions to get the full picture of the customer's real journey in a single interaction, in order to identify and improve any channels that failed to fulfill their requirements. Improving self-service optimization is often a quick win that can provide immediate economic benefit to businesses: in the UK, a mean average of 9% of calls that go into an IVR system are 'zeroed-out' - rejected by the customer in favour of an operator - and in the US, a staggering 26% fail the self-service test. Businesses using speech analytics to review these failed self-service sessions will be able to categorise many of them in order to improve the processes at a macro-level. Common findings from the analysis of these calls is that the IVR system was poorly worded or menu choices are not intuitive or match current service choices. Other failures occur through mistakes in IVR routing, and there may also be problems with a lack of customer awareness that various activities can be carried out by self-service. INCREASING PROFITABILITY Debt collection and improving cross-selling & up-selling Although many debt collection firms have detailed scripts for their agents - often driven by the need to comply with regulations - the results, such as the promise-to-pay ratio - can differ widely by agent. Speech analytics provides two benefits for debt collectors: the ability to prove compliance, and through the analysis of successful and unsuccessful calls, the chance to understand the type of agent language and behaviour that yields the best results, and share these with underperforming agents. The same principle of matching successful outcomes with particular call traits can be used for improving cropsselling and up-selling rates in sales environments. Managing customers at risk of churn Using real-time analytics, linked with a company's own CRM systems, agents can be provided with up-to-thesecond advice on how to handle customers identified as being at risk of churn, including linking what the customer is saying on the call back to the transactional model in order to update the best offer available for that customer. Feedback on marketing campaigns Tracking customer comments and outcomes after the advent of a marketing campaign can mean the difference between success and failure. Messages that are incorrectly understood can be identified and altered quickly before the contact centre becomes swamped with calls about the issue. 19

20 Phone-based contracts Real-time speech analytics mean that phone-based contracts can be seen to be completed first-time, with all relevant information provided to the customer on the call, and red-flagged on the agent's screen if they have missed saying anything vital, or made an error. This reduces the need to call a customer back and avoids any dispute over whether a legitimate contract has been made. 20

21 BUILDING A BUSINESS CASE FOR SPEECH ANALYTICS There is no generally-agreed job role that initially identifies a potential requirement for speech analytics. Some vendors state that it is the commercial side that starts a conversation, with IT people taking over. Others say that it is the contact centre, the legal department or the QA/QM teams who show most interest. As speech analytics can be used to improve business intelligence, increase contact centre efficiency, improve agent quality and performance, execute compliance and optimize business processes throughout the organization, it is little wonder that there is no well-worn path to the vendors' doors. However, all vendors agree that in their experience, having a senior and empowered contact within the business who knows what they want to achieve through speech analytics is vital to the success of the project. It should also be noted, that postimplementation, well-trained and empowered supervisors, trainers and coaches are key to getting and maintaining a positive outcome from the use of the solution. End-user question: "Any tips for how best to engage the business, and inspire them to make changes?" (various) The only answer to this is: try it! A proof of concept trial is the best way of providing justification to the business. The key here is to find a solution with an easy and fast set up so that any cost of trial is minimal. Call Analyzer is so automated, due to the inclusion of a specific automated configurator tool, that a trial can be set up within a few days and allowed to run for a couple of weeks to provide the data needed for business justification. The costs of which can be absorbed into the overall price of the solution when purchased. 21

22 ESTIMATING ROI Return on investment for speech analytics can come from numerous sources, depending upon how the solution is used. Generally, it will come from the avoidance of a specific cost, (including the reduction of a risk in the case of compliance), or the increase in revenue. The return on investment of speech analytics used for compliance can at first glance be difficult to prove, but it is the avoidance or reduction in litigation and regulatory fines which can be placed against the cost of the solution. Large banks will have funds put away running into the tens of millions of pounds each year against the possibility of paying out, and any significant reduction in fines would pay for a speech analytics solution very quickly. In the UK, the banking industry has put aside several billion pounds to pay compensation for the mis-selling of PPI (payment protection insurance), and having the ability to prove that no regulations had been broken would have been of great use. Most vendors have tools which can be used to estimate return on investment, often based on what they have seen in similar operations elsewhere, and they are keen to share them with potential customers. Vendors' own estimates of the time taken for the solution to pay for itself vary between 6 and 18 months. End-user comment: "The biggest question I have is how to build a strong enough Return on Investment (ROI) to get corporate buy in to purchase speech analytics software." (Various) Building a strong ROI needs a multi-pronged approach: a) Firstly you need to assess how much time and money is currently spent on evaluating calls. More often than not people monitoring calls will only have time to listen to a random selection of calls, which means they could miss vital information, e.g. customer issues, poor performing agents or abusive calls. The alternative is to spend hours listening to entire calls on the off chance they will get valuable feedback on their agents and customers which is costly. In comparison using live speech analytics means that no listening is required as solutions such as Call Analyzer record, analyse and report on every call in real-time therefore minimising evaluation overheads and reporting on every call. b) If your organisation is bound by regulations such as FSA rules, you are vulnerable to large financial penalties as well as customers requesting their money back if they have been mis-sold products and services. Using Call Analyzer you can ensure that key, mandatory phrases and contract terms are said on every call. If your organisation is an outsourced call centre this will provide competitive advantage i.e. you can say to every client that you monitor 100% of calls and have the ability to affect live calls as they occur. Ask yourself, what value do you place on getting calls wrong and what are the financial risks and consequences of non-compliance? c) Using live call analysis can reduce your evaluation head count by well in excess of 50%. This resource can be reallocated as extra call agent headcount or more efficient supervisory roles. 22

23 Variables to be considered for ROI measurements include: Cost reduction: Reduction in headcount from automation of call monitoring and compliance checking Avoidance of fines and damages for non-compliance Reduction in cost of unnecessary callbacks after improving first-call resolution rates Avoidance of live calls that can be handled by better IVR or website self-service Reduced cost of QA and QM Lower cost per call through shortened handle times and fewer transfers Lower new staff attrition rates and recruitment costs through early identification of specific training requirements Revenue increase: Increase in sales conversion rates and values based on dissemination of best practice Increase in promise-to-pay ratios (debt collection) Optimised marketing messages through instant customer evaluation Reduced customer churn through dynamic screen-pop and real-time analytics Quicker response to new competitor and pricing information Also, the improved quality of agents, better complaints handling and improved business processes outside the contact centre should be considered. Against these potential positives, costs to consider include: Licence fees or cost per call analysed IT costs to implement (internal and external) Upgrade to call recording environment if required Bandwidth if hosted offsite: the recording of calls is usually done on a customer's site, so if the speech analytics solution is to be hosted, it will involve of lot of bandwidth, which will be an additional cost, especially when considering any redundancy Maintenance and support agreements, which may be 15-20% annually of the original licencing cost Additional users - headcount cost - decide who will own and use it, do you need a speech analyst, etc. Extra hardware e.g. servers Ongoing and additional training costs if not included Extra work generated by findings May need extra software to extract data from the call recording production environment. 23

24 INHIBITORS TO SPEECH ANALYTICS A major inhibitor to uptake is an awareness within the company that their environment is not yet ready for speech analytics, in that they may still not have a reliable recording environment or an optimized QM or QA process. End-user comment: "Speech can get sidelined, and a lot of the justification for this is that speech analytics works best on the ideal platform (tonnes of metadata from a telephony set up with cradle to grave info regarding each and every contact, all calls recorded in stereo with no file corruption and everything attributed to the correct agent/team/dept via integration with a well maintained workforce management system). So speech ends up waiting at the back of the line until the Goldilocks platform is in place before any major investment such as on-site transcription servers gets put its way." (Major global retailer). The best speech analytics solutions provide an easy to use and understand complete package which does not require an assessment of current telephony and call recording set up within the customer s organisation. Call Analyzer in particular provides a solution which does not interfere with existing equipment and requires only a WAN connection to operate. By incorporating the following features the customer can be assured that they have the best quality data from which to analyse: Dual channel recording (otherwise known as speaker separation) Digital recording and playback to the quality of the line A live feed which is unencrypted when analysed and delivered in an uncompressed state during analysis All of which help guarantee the best results. 24

25 Security and regulation was also mentioned on several occasions as a potential issue. End-user question: "Does PCI prevent the use of speech analytics?" (UK outsourcer) The simple answer is no. On the contrary GemaTech uses live speech analytics to trigger PCI compliance, i.e. when the call agent asks for the customer s long credit card number, saying a pre-determined phrase triggers the PCI compliance feature which then obfuscates the voice within the call recording making it fully compliant and without interfering with the call or user experience in any way, and which is also stopped when another predetermined phrase is said by the agent once the customer has given their details. Some businesses consider that their existing call recording and manual quality monitoring processes are sufficient, and fail to understand the potential business value of speech analytics. 25

26 THE IMPLEMENTATION AND USE OF SPEECH ANALYTICS PRE-IMPLEMENTATION: SELECTING A SOLUTION Initiators and the Project Champion Lots of budget is held with marketing, website or customer experience teams, rather than at contact centre level, and these teams are now seeing that the contact centre is a big part of people's experience of dealing with a company. To get the most from a speech analytics solution, especially the more complex systems, businesses need to identify and empower a senior project champion, overseeing a cross-functional team. The champion must have a strategic view of what analytics can provide, as well as being able to understand the operational and technical requirements of the contact centre and IT teams. Some milestones for selecting a vendor include: Identify interested cross-functional parties in the organisation and get a senior project champion Choose a specific area of improvement and benchmark it (baseline analysis). This may be something to consider in trial mode - as it is manageable, quick to identify, not reliant on other elements or affecting them, so a fair before & after measurement is possible. Input from relevant departments into deliverables, explaining and agreeing what they have to put into this themselves Create a vendor longlist and have informal discussions with them Consider technical constraints and internal cultural preferences (e.g. propensity to host vs CPE) and build vendor shortlist / request for proposal Selection, including their ability to build an ROI model / proof-of-concept trial for you, plus referenceable sites if required. Check interoperability and willingness to work between incumbent recording vendors and new speech analytics vendors. Reference sites using same combination of vendors recommended if possible Deployment either as trial or full roll-out. 26